Improving causal interpretation in ultra-processed food and cardiovascular disease research: A systematic review of quantitative methods — The Association Specialists
  1. Schedule
  2. Oral Presentations - Advancing Epidemiological Methods 1
  3. Improving causal interpretation in ultra-processed food and cardiovascular disease research: A systematic review of quantitative methods

Improving causal interpretation in ultra-processed food and cardiovascular disease research: A systematic review of quantitative methods (24170)

Tefera Mekonnen 1 , Zegeye Abebe 1 , Anteneh Mengist Dessie 1 , Teketo Tegegne 2 , Tolassa Ushula 2 , Kacie Dickinson 1 , Zumin Shi 3 , Robert Adams 1 , Jason Wu 4 , Mario Siervo 5 , Yohannes Adama Melaku 1
  1. Flinders University, Adelaide, Australia
  2. Deakin University, Melbourne, Australia
  3. Qatar University, Doha, Qatar
  4. UNSW, Sydney, Australia
  5. Curtin University, Perth, Australia

Background: Evidence linking ultra-processed food (UPF) intake with cardiovascular disease (CVD) has expanded rapidly. However, differences in how UPF exposure, outcomes and covariates are measured and analysed may affect the strength, consistency and causal interpretation of this evidence. We reviewed quantitative methods used in UPF–CVD studies, focusing on features relevant to causal inference.

Methods: We systematically reviewed adult studies examining UPF consumption and CVD outcomes. After dual independent screening of 13,055 records, 45 studies published between 2017 and January 2025 were included: 42 prospective cohorts, two modelling studies and one controlled quasi-experimental intervention, representing 4,332,932 adults aged 18–79 years across all inhabited continents. We extracted information on study design, dietary assessment, UPF classification, exposure specification, outcome ascertainment, statistical models, covariate selection, sensitivity analyses and causal inference methods.

Results: Most studies assessed diet using validated food frequency questionnaires, while fewer used 24-hour recalls; more than half relied on baseline dietary assessment only. Although all studies applied the NOVA framework, reporting of food classification procedures was limited, with only about one-quarter describing classifier number or expertise. UPF exposure was expressed inconsistently as grams, percentage of energy intake, servings or frequency, with variable treatment of alcohol and limited reporting of water handling in weight-based measures. CVD outcome definitions also differed, particularly regarding fatal and non-fatal events. Conventional regression dominated while causal inference methods were rarely used as the primary framework. Majority studies adjusted for variables that may lie on the causal pathway (mediators) raising concerns about overadjustment. Although higher UPF intake was generally associated with increased CVD risk, estimates varied across studies.

Conclusions: UPF–CVD research is limited by heterogeneous exposure definitions, inconsistent outcome ascertainment, variable covariate adjustment and limited use of causal inference frameworks. Future studies should use target trial thinking and principled confounder selection to strengthen causal interpretation.